Click here to load reader

TAX AVOIDANCE, EVASION, AND ADMINISTRATION · TAX AVOIDANCE, EVASION, AND ADMINISTRATION Joel Slemrod The University of Michigan Shlomo Yitzhaki The Hebrew University of Jerusalem

  • View

  • Download

Embed Size (px)



    Joel SlemrodThe University of Michigan

    Shlomo YitzhakiThe Hebrew University of Jerusalem

    November 5, 1999

    Forthcoming in the Handbook of Public Economics, edited by Alan Auerbach and MartinFeldstein. We are grateful to Wojciech Kopczuk for valuable research assistance.Helpful comments on an earlier draft were received from Jim Alm, Alan Auerbach, LenBurman, Brian Erard, Firouz Gahvari, Roger Gordon, Jim Hines, Jonathan Kesselman,Louis Kaplow, Agnar Sandmo, Dan Shaviro, Eric Toder, Gideon Yaniv and theparticipants at the NYU Law School Colloquium on Tax Policy and Public Finance.

  • Tax Avoidance, Evasion, and Administration

    Joel Slemrod and Shlomo Yitzhaki


    1.a. Why Avoidance, Evasion and Administration are Central, NotPeripheral, Concepts in Public Finance

    Most economic analysis of taxation presumes that tax liability can be ascertained

    and collected costlessly. As a description of reality, this is patently untrue. For example,

    in the U.S. the Internal Revenue Service (henceforth IRS) estimates that about 17% of

    income tax liability is not paid;1 the figure for most other countries is probably higher.

    Furthermore, the resource cost of collecting what is paid can be large, in the U.S.

    probably about 10% of tax collections.2 The tax structures themselves are undoubtedly

    skewed by the realities of tax evasion, avoidance, and administrative costs.

    The standard models of taxation and their conclusions, need to be modified in the

    light of these realities. Many practitioners of tax advice believe that this change in

    emphasis is essential; for example, Casanegra de Jantscher (1990, p. 179) goes so far as

    to say that, in developing countries, "tax administration is tax policy."3 Bird (1983),

    Mansfield (1988), and Tanzi and Pellechio (1996) are useful summaries of the practical

    problems of the interaction of tax policy and tax administration.

    1 Internal Revenue Service (1996).2 Slemrod (1996a).3 Others disagree. Groves (1974, p. 25) offers that

    "Vetoing tax measures because of the difficulty of administering them is in most casesless compelling than doing so on the ground of their failure to conform to acceptableprinciples. Administration is usually amenable to improvement where violation of firstprinciples is not. And administration of a given tax may often be improved mosteffectively in the process of attempting to administer it. The point is sometimes crucial inrecommending taxes for so-called underdeveloped economies in our own time."

  • 2

    1.b. The Evolution of Tax Structures

    Scholars of the historical evolution of tax structure, notably Hinrichs (1966) and

    Musgrave (1969), have also stressed the importance of tax administration issues. They

    note that modern tax structure development has generally been characterized by a shift

    from excise, customs, and property taxes to corporate income and progressive individual

    income taxes.4 This shift has been made possible by the expansion of the market sector

    and relative decline of the rural sector, the concentration of employment in larger

    establishments, and the growing literacy of the population. Further changes in the

    technology of tax administration, including globalization and financial innovation, may

    now be pushing us away from progressive income taxes toward tax systems that rely

    more on broad-based consumption taxes such as the value-added tax (VAT), flatter rate

    structures for income taxation, or the "dual income tax" system recently adopted by

    certain Scandinavian countries, and described in Sorensen (1994).

    Alt’s (1983) treatment of the evolution of tax structure stresses the role of

    administrative and compliance costs. He argues that it has become increasingly easy to

    collect taxes from organized business rather than from households, and that one

    explanation for the widespread adoption of the VAT is that it imposes compliance costs

    without raising administrative costs, through incentives for self-policing. Kau and Rubin

    (1981) focus on changes in the cost of collecting taxes, and successfully relate growth of

    the U.S. federal government to reasonable correlates of collection cost, such as the

    literacy rate, the extent of female labor force participation, and the extent of the

    4 Although Hinrichs (1966) points out that tax structure development began with direct taxes rather thanindirect taxes.

  • 3

    agricultural sector. Balke and Gardner (1991) contend that declining marginal collection

    costs can explain the stepwise growth in the size of government and the changes of

    taxation observed in the U.S. and U.K. They argue that major wars coincide with

    permanent improvements in tax instruments and tax collection technology, which

    facilitated permanent expansions in government size thereafter.

    Putting aside the role of administrative issues in explaining the evolution of tax

    levels and tax structures, it is indisputable that these considerations are critical

    determinants of tax policy at a point in time. For example, an important set of generic

    aspects of income tax structure, such as the absence of taxation of imputed rents from

    consumer durables, taxation of capital gains (if at all) on a realization basis, and pre-set

    depreciation schedules, are undoubtedly largely driven by practical concerns of

    administerability. For these reasons, we believe that consideration of evasion, avoidance,

    and administration is essential to the positive and normative analysis of taxation. Our

    view corresponds closely to that of Blough (1952, p. 146):

    It is tax policy in action, not simply the wording of the statute, thatdetermines how much the taxpayer must pay, and the effects of thepayment. Knowledge of the statute is only a start in knowing a taxsystem. The interpretations placed on language by administratorsand courts, the simplicity and understandability of tax forms, thecompetence and completeness of audit, the vigor and impartialityof enforcement, and the promptness and finality of action allinfluence the amount of revenue collected, the distribution of thetax load, and the economic effects of the tax.

    In this chapter we organize, explicate, and evaluate the modern literature that

    incorporates these considerations into the economics of taxation.5 We do not claim to

    5 Other surveys of these issues include Cowell (1990b), Andreoni, Erard, and Feinstein (1998), Roth,Scholz, and Witte (1989), and Alm (1999). Because of space constraints we have omitted any discussion ofthe role of tax practitioners, corruption in tax administration, tax amnesties, or business tax evasion.

  • 4

    have put together a comprehensive survey of this literature, which is huge and multi-

    faceted, rather a guide to what we feel are the most important issues and contributions in

    this area.

    1.c. Evasion, Avoidance, and Real Substitution Response

    We begin with some definitions. The classic distinction between avoidance and

    evasion is due to Oliver Wendell Holmes, who wrote

    "When the law draws a line, a case is on one side of it or the other,and if on the safe side is none the worse legally that a party hasavailed himself to the full of what the law permits. When an act iscondemned as evasion, what is meant is that it is on the wrong sideof the line..." (Bullen v. Wisconsin (1916), 240. U.S. 625 at p.630).

    Thus, the distinguishing characteristic of evasion is illegality.6 In practice, of

    course, there are many gray areas where the dividing line is not clear, and sometimes the

    tax authorities may inappropriately characterize particular cases.

    One can draw a further distinction within the class of legal responses to taxation.

    At times we will refer to real substitution responses, or real responses for short, as those

    responses which come about because the tax law changes the relative price of different

    activities, and that induce taxpayers to respond by choosing a different consumption


    Conceptually distinct from real substitution responses are efforts to reduce one’s

    tax liability without altering one’s consumption basket, which we will refer to as

    avoidance. These are actions taken in response to the tax system that do not involve shifts

    6 Kay (1980, p. 136) offers a different pair of definitions for evasion and avoidance: "Evasion is concernedwith concealing or misrepresenting the nature of a transaction; when avoidance takes place the facts of thetransaction are admitted but they have been arranged in such a way that the resulting tax treatment differsfrom that intended by the relevant legislation."

  • 5

    along a given budget set. This definition covers a broad range of behaviors. One

    example is to pay a tax professional to alert one to the tax deductibility of activities

    already undertaken. Another example is to change the legal form of a given behavior,

    such as reorganizing a business from a C corporation to an S corporation,7

    recharacterizing ordinary income as capital gain, or renaming a consumer loan as a home

    equity loan. A third example is tax arbitrage, when economically equivalent, but

    differentially-taxed, positions are held simultaneously long and short, producing tax

    savings. Finally, retiming a transaction to alter the tax year it falls under is an example of


    Fine distinction among the types of behavioral response to taxation is not possible

    and is for many issues not crucial. In general, changes in the tax structure will induce all

    the different kinds of response. Indeed, one of the goals of this chapter is to emphasize

    the common analytical aspects of issues that have traditionally been kept distinct.

    1.d. General Framework

    Although there may be reasons, discussed later, for distinguishing among these

    categories of response to taxation, there is a common framework for analyzing these

    issues. Given the structure of the tax system and enforcement process, taxpayers are

    faced with opportunities to reduce their tax payments, or expected tax payments. There is

    a private cost to taking advantage of these opportunities, which may take the form of an

    altered consumption basket, an increasing probability of detection of, and penalty for,

    evasion, and/or a real resource cost of effecting avoidance or concealing evasion. This

    7 Under U. S. tax law an S Corporation retains the legal characteristics of a corporation, but is taxed as apass-through entity such as a partnership. There are restrictions to becoming an S corporation, most notablyon the maximum number of shareholders.

  • 6

    private cost depends on policies of the government that include, but are not limited to, the

    setting of tax rates and bases. The parameters of the tax administration and enforcement

    policies also matter, but these policies themselves are usually costly.

    The tax system establishes the relative prices among this broad set of taxpayer

    activities. In the standard model, it establishes the relative price of leisure and other

    goods, as well as the relative price among the set of goods. In a more general framework

    it also sets the price of "honesty," meaning the incentives to evade, and establishes the

    cost and reward to legally reducing taxes via avoidance activities. The dimensions of

    taxpayer response interact. For example, real behavior may alter the cost of avoidance or

    evasion, thus changing the effective prices of real activities.

    Although these are common themes, the literature to date has tended to isolate

    pieces of the overall problem. We follow that practice here, by beginning in Section 2

    with a discussion of the now standard economic model of tax evasion.8 Then in Section 3

    we introduce models that apply more generally to both evasion and avoidance. We then

    look at the empirical evidence, first in Section 4 about evasion, and then in Section 5 on

    avoidance. The remainder of the chapter addresses the implication for tax analysis of

    introducing these issues. Section 6 examines the fundamental issues of positive tax

    analysis, while Section 7 addresses normative issues. Section 8 concludes.

    8 There is a vast literature which investigates non-economic perspectives on tax evasion. We do not havethe space to discuss or evaluate this literature, and refer the interested reader to Roth, Scholz, and Witte(1989).

  • 7


    2.a. The Allingham-Sandmo-Yitzhaki Model

    Suppose that the true tax base is known to the taxpayer, but is not costlessly

    observable by the tax collection agency. Then, under certain circumstances, the taxpayer

    may be tempted to report a taxable income below the true value. In the seminal

    formulation of Allingham and Sandmo (henceforth A-S) (1972),9 what might deter an

    individual from income tax evasion is a fixed probability (p) that any taxable income

    understatement will be detected and subjected to a proportional penalty (θ) over and

    above payment of the true tax liability itself. Later we introduce and discuss at length a

    "technology of evasion," in which evasion involves costs to the evader that might depend

    on the income and the amount of tax evaded.

    In the A-S model, all real decisions, and therefore taxable income (y), are held

    fixed; only the taxpayer’s report is chosen. The risk-averse taxpayer chooses a report (x),

    and thus an amount of unreported income y-x, in order to maximize expected utility:

    (1) EU = (1 - p)U(ν + t(y - x)) + pU(ν - θ(y - x)),

    where ν is true after-tax income, y(1 - t), t being the rate of (proportional) income tax.

    The von Neumann-Morgenstern utility function U(⋅) represents the individual’s

    preferences toward risk. In this model the choice of whether and how much to evade is

    akin to a choice of whether and how much to gamble. Each dollar of taxable income

    understatement offers a payoff of t with probability (1-p), along with a penalty of θ with

    9 The Allingham-Sandmo paper applies to tax evasion the approach of the classic paper on the economics ofcrime by Becker (1968).

  • 8

    probability p. If and only if the expected payoff to this gamble, (1 - p)t-pθ, is positive,

    every risk-averse taxpayer will chance some evasion, with the amount depending on the

    expected payoff and the taxpayer’s risk preferences.

    A critical issue, pointed out by Yitzhaki (1974), is whether the penalty for

    discovered evasion depends on the income understatement, as A-S assume, or on the tax

    understatement, as more accurately reflects practice in many countries. In the latter case,

    the maximand becomes (1 - p)U(ν + t(y - x)) + pU(ν - θt(y - x)), and the expected payoff

    per dollar of evaded income becomes (1 - p)t - pθt. This is an important change, because

    it means that the tax rate has no effect on the terms of the tax evasion gamble; as t rises,

    the reward from a successful understatement of a dollar rises, but the cost of a detected

    understatement rises proportionately. The first-order condition for optimal evasion


    (2)U yAU yU


    ' ( )

    ' ( )( )= −1


    where yA and Uy refer to net income in the audited and unaudited states of the world,

    respectively. Note that t does not appear in equation (2), other than via an income effect

    in the definition of yA and Uy . Compare this to the original A-S formulation, where t

    would be a multiplicative factor in the denominator of the right-hand side, implying that

    increases in t would proportionally increase the reward to getting away with understating

    income, but not proportionally increase the penalty, making evasion more attractive.

    Regardless of whether the penalty depends on the tax understatement or income

    understatement, more risk-averse individuals will, ceteris paribus, evade less.

    Individuals with higher income will evade more as long as absolute risk aversion is

  • 9

    decreasing; whether higher-income individuals will evade more, as a fraction of income,

    depends on relative risk aversion. Evasion relative to income will decrease, increase or

    stay unchanged as a fraction of income depending on whether relative risk aversion is an

    increasing, decreasing, or constant function of income. Increases in either p or θ will

    decrease evasion.

    Increasing t has both an income effect and, possibly, a substitution effect. If the

    taxpayer has decreasing absolute risk aversion, the income decline makes a less risky

    position optimal. An increase in t has a substitution effect, increasing the relative price of

    consumption in the audited state of the world, and thereby encouraging evasion, if the

    penalty is related to income, rather than tax avoided. In the latter case, if the penalty is

    related to the tax evaded, a tax increase has no substitution effect, so that an increase in t

    reduces evasion as long as there is decreasing relative risk aversion.10

    This simple version of the A-S model has been criticized on the grounds that it

    fails a simple reality check. If p is the fraction of returns audited in the U.S., about 0.015,

    and θ is the statutory penalty for non-criminal evasion, about 0.2, then based on the

    degree of risk aversion exhibited in other situations people should be evading a lot more

    than they apparently do. The intriguing question becomes why people pay taxes rather

    than why people evade. Much subsequent research, some of it surveyed below, has been

    addressed to reconciling the facts with the theory.

    10 Note the similarity to the standard model of the effect of taxation on the optimal portfolio, in which a taxincrease can increase the demand for the risky asset (Domar and Musgrave, 1944 ). One difference is that,in a portfolio model, it is arguably inappropriate to ignore the effect of the tax scheme on the variability ofgovernment revenues (Gordon and Wilson, 1989). This issue can be sidestepped in the context of a taxevasion model, because the "risks" are independent and therefore there is no social risk involved.It is important to distinguish the effect of a change in the environment on evaded income (y - x) versus theimpact on evaded tax liability, t(y - x). With respect to changes in p and θ, there will be no interestingdistinction. However, when t increases it is certainly possible that (y - x) may decline at the same time t(y -x) increases.

  • 10

    In the A-S model what limits the amount of evasion attempted is the taxpayer’s

    risk aversion. At some point further evasion becomes just too big a gamble, so that at the

    chosen amount of evasion the marginal gain in expected tax savings is exactly offset by

    the marginal disutility of the extra risk taken on.11 The model also predicts that a risk-

    neutral individual would either remit no tax at all, or do no evasion, depending on

    whether evasion has a positive expected payoff. This "either-or" prediction is eliminated

    if the probability of detection is an increasing function of the amount of evasion, which is

    likely to characterize most tax systems. The implications of introducing an endogenous p

    depend on the precise relationship between p and evasion. For example, consider the

    case (discussed in Yitzhaki (1987)) where p is an increasing function of evaded income

    (y - x). The risk-neutral taxpayer chooses x to maximize expected income,

    (3) EY = ((1-p[y - x])(ν + s) + p[y - x](ν - θs)),

    where s ≡ t(y - x) is understated tax. If p' ≡ ∂ ∂p y x/ ( )− is positive, the first-order

    condition becomes

    (4) 1 1− − = +p p p s tθ θ' ( )( / ) .

    In this case, evasion will be constrained by the fact that p increases to offset what would

    otherwise be an increase in expected income.

    The either-or prediction in the case of a risk-neutral taxpayer is also eliminated if

    there are distinct sources of income, each of which is subject to its own p. For example,

    11 In the language of Kolm (1973), the evasion is accomplished by "the mere stroke of a pen." We considerbelow where the evasion is facilitated by supplying labor to an "underground" sector which offers betterconcealment possibilities.

  • 11

    employee labor income has a high p (due to information reporting by employers and

    computer matching), while "moonlighting" income has a much lower p. Faced with this

    situation, a risk-neutral individual reports all or none of each of the several sources of

    income, but may certainly report a fraction of total income.12

    The endogenous probability of detection can of course be applied to the case of a

    risk-averse taxpayer. In this case, at the margin the gain in expected value is offset by a

    combination of increased risk-bearing and an increased probability of detection. Cremer

    and Gahvari (1994) generalize this notion by introducing what they call a "concealment

    technology," which in our notation takes the form p(y - x, ( )( ) m,y/xy − ) where m

    represents taxpayer expenditure on concealment. The notion that the probability of

    detection can be increased by the taxpayer's expenditure is also present in Usher (1986),

    Kaplow (1990), Cowell (1990), and Mayshar (1991).

    2.b. Jointness With Labor Supply

    Of particular interest is the relationship between the tax report decision and other

    consumer decisions. Most attention has been paid to labor supply, where the individual

    chooses how much labor to supply and how much labor income to report. The decision

    about how much income to report is made simultaneously with the decision of how much

    to work, so that it is impossible to adjust labor supply based on whether one is caught

    evading. This problem may be posed as how much of a homogeneous labor income to

    report, which is equivalent to simultaneously choosing one's consumption basket and

    12 As discussed in the next section, differential detection rates could also affect the sectoral supply of labor.

  • 12

    exposure to risk.13 Models that belong to this group are based on extensions of the A-S

    model. Alternatively, the problem may be posed in the context of a model of the

    underground economy, in which there are two sectors with possibly different equilibrium

    wage rates and other different circumstances. The latter class of models allows for wage

    adjustment in response to policy changes, and thus are general equilibrium in nature.

    In the extensions of the A-S model, the first-order condition for labor supply

    differs from that in a model without tax evasion only in that it contains mean marginal,

    instead of marginal, utilities. Whether mean marginal utility is bigger or smaller than the

    marginal utility depends on the sign of the second derivative of marginal utility, which is

    the sign of the third derivative of the utility function. On top of that, if utility is non-

    separable, the marginal utility functions depend on the sign of cross-derivatives, which

    further complicates the problem.14 Baldry (1979) and Pencavel (1979) stress the

    difficulty of reaching any clear-cut comparative statics conclusions from such a model;

    the response of reported income to changes in tax rates, penalties, and fines becomes

    ambiguous. Thus, most models are based on particular restrictive assumptions about the

    utility function. For example, if the utility function is separable in consumption and

    leisure, then the marginal utility of leisure is independent of consumption. If, in addition,

    the marginal utility of consumption is linear (as in the function

    U C L C C L( , ) = + + +α β γ δ2 ), the first-order condition for optimal labor supply is

    (5) ],L[U]s)p1(wL[wU)t1( 21 =θ−+−

    13 Models of this type resemble models of choice among risky occupations (e.g., Kanbur, 1979), except thatin the latter the occupational choice is usually discrete, so that a "diversified" occupational portfolio is notallowed.14 Moreover, the conditions that the equilibrium investigated is on the increasing portion of the Laffer curvealso depend on the curvature of the third derivative of the utility function, further complicating the issue.

  • 13

    where s is the tax evaded and (1-pθ)s is the expected gain from evasion. Because evasion

    increases expected consumption for any given amount of leisure without changing the

    real wage, leisure would increase, and labor supply would decline. The real wage does

    not decline because the evasion opportunities are independent of the amount of work

    done. The critical importance of the relationship between the real consumption choices

    and the evasion or avoidance opportunities comes up again in the more general models

    discussed in Section 3. There we discuss cases where the avoidance opportunities do

    affect the real wage. In situations where labor income in the formal sector is reported by

    the employer to the tax enforcement agency as a matter of course, the only way to evade

    tax may be by "moonlighting"-- working extra hours at a different job -- or by switching

    completely to the informal sector or "underground economy."

    2.c. Other Uncertainty

    The basic model can also be extended to deal with other sources of uncertainty.

    Andreoni (1992) introduces a temporal nature to the tax evasion decision, recognizing the

    fact that the penalty for tax evasion, if detected, is assessed later than the tax saving.

    Andreoni deviates from the majority of the literature which assumes efficient market

    environments, and instead assumes that the taxpayer is constrained by credit rationing.

    Due to uncertainty, the income of the taxpayer fluctuates, as does the shadow price of

    income. Provided that non-monetary punishments are high enough to deter one from

    non-repayment of penalties and tax evaded, evasion may be viewed as a way of

    "borrowing" from the IRS. A constrained taxpayer may find it optimal to borrow when

    the shadow price of money is high enough during evasion and relatively low during

  • 14

    repayment.15 Andreoni models a situation where, in bad times, individuals evade as a

    way to smooth income streams; thus the IRS is a "loan shark." The conditional

    repayment of the loan occurs in a better state of the world.

    Another aspect of uncertainty concerns the unpredictability of the tax liability

    itself, which arises when the "correct" tax liability is not clearly defined.16 Uncertainty

    of true tax liability can be modeled by extending the Allingham and Sandmo framework.

    Scotchmer and Slemrod (1989) construct a model where, upon audit, the assessed tax

    liability is symmetrically centered around a known value with an equal probability of one-

    half. In this case the very concept of income understatement becomes problematic

    because the taxpayer is uncertain whether any given income declaration is correct or not.

    There are now three possible outcomes that the taxpayer must consider. If the

    return is not audited (with probability 1 - p), true taxable income is irrelevant--the

    taxpayer merely pays the tax due on his declared taxable income. If the return is audited,

    there are two possible outcomes, depending on what the assessed tax liability turns out to

    be. Scotchmer and Slemrod (1989) show that increasing the dispersion of possible

    assessed taxable incomes induces increased compliance, given weak conditions about the

    taxpayer’s attitudes toward risk. The intuition is that, for a given reported income, more

    dispersion lowers income in the least desirable state of the world, when the taxpayer is

    audited and his taxable income is determined to be the highest possible value. This

    increases the marginal utility of income in that state of the world, which is accomplished

    15 In this case the government may find it optimal to encourage tax evasion. The optimality of such a policydepends crucially on the non-existence of alternative methods of borrowing, including negotiated paymentterms with the IRS, which can in some situations be arranged in the U.S.16 Long (1981) argues that the IRS exploits the unpredictability of tax liability to enhance its powers byusing it as a license to decide cases in whatever way serves the government’s interest at the time. She alsonotes that unpredictability makes the IRS’s burden in providing criminal intent (rather than inadvertenterrors) more difficult.

  • 15

    by increasing reported income and thus subjecting oneself to a lower penalty in the event

    this state of the world occurs. As long as the taxpayer exhibits declining absolute risk

    aversion, increasing the report is the optimal response.

    Beck and Jung (1987a) show that this conclusion may not hold when there is a

    continuous range of possible taxable income assessments. In this case one marginal

    benefit of increasing the income report is that it reduces the probability that a fine will be

    assessed. For a taxpayer reporting income below the mean of possible assessment, an

    increased dispersion of possible assessed incomes decreases the likelihood that the

    income report will be declared insufficient and a fine assessed, so that this component of

    marginal benefit is reduced. Thus, it is theoretically possible that increased dispersion

    will cause a lower report.

    Note that uncertainty does not reduce tax evasion by as much as it reduces

    aggregate noncompliance in the sense of true aggregate tax liability minus tax paid. This

    is because one effect of uncertainty is to induce some taxpayers to pay more tax than they

    are legally obligated to pay, which reduces aggregate noncompliance but not the amount

    of individual tax evasion.

    Scotchmer (1989) allows for the possibility that, by expending resources, the

    taxpayer can reduce the uncertainty of tax liability. The resources can be in the form of

    research by the taxpayer himself, or in the form of professional assistance hired. In this

    case the cost of unpredictability includes not only the disutility caused by uncertain tax

    liability but also the resources expended to reduce the uncertainty.

  • 16

    2.d. General Equilibrium Considerations

    The A-S model and its direct descendants address only the demand for tax evasion

    by (potential) taxpayers. One might also consider the "supply" of evasion, and ponder the

    general equilibrium considerations of demand having to equal supply.

    One context for this extension is the underground economy. Kesselman (1989)

    develops a set of models in which there are two sectors -- above-ground and

    underground--which produce two distinct goods. Workers are homogeneous in their

    gross productivity in each sector of the economy (and in their consumption preferences),

    but must work only in one sector or the other. The workers, though, have differential

    distaste and risk aversion for tax evasion, and differential efficiency in concealment and

    other skills needed to operate successfully in the underground economy.

    Although the precise results are model-dependent, three general conclusions

    obtain: (i) much of the gain from evasion may be shifted from the evaders to the

    consumers of output through lower prices, and the "marginal" evader gains nothing; (ii)

    relative price effects tend to dampen the impact of tax rate changes on the extent of

    evasion, and (iii) the effects of evasion on the marginal revenue response to tax rate

    changes will depend on consumers’ elasticity of substitution between the sectoral outputs.

    A key aspect of the foregoing model is that the act of tax evasion is tightly tied to

    the production of a distinct good. This need not be true, as is indicated by the

    simultaneous presence of above-ground and underground housepainters, repair people,

    and so on. Still, there is certainly evidence that evasion is concentrated in particular

    sectors, such as those that supply services directly to homeowners, because of the small

  • 17

    scale of production that can aid concealment and the lesser need for receipts compared to

    services provided to businesses.


    Because Allingham and Sandmo addressed tax evasion as a gamble, much of the

    subsequent literature focused on models in which taxpayers' risk aversion, and therefore

    higher-order characteristics of utility functions, played an important role. This focus has

    to some extent obscured other important aspects of the issue, such as the tax concealment

    technology, and also obscured the common aspects of evasion and what we have called

    avoidance. To highlight these issues we turn now to more general models of behavioral

    response to taxation.

    Mayshar (1991) poses the taxpayer’s problem as:

    (6) Max U(Y,L) X,S,L,Y

    subject to X=w[L-S-m(E)]

    and Y=X-T(X,S,E),

    where X is output, S is sheltering effort, L is total labor effort, and Y is consumption.

    Mayshar labels T(⋅) the "tax technology;" it specifies the maximal taxes, T, collectible

    from a base X, when the tax authority selects a vector E of policy instruments, while the

    taxpayer devotes S in labor units to sheltering activity. It is reasonable to assume that

    TX>0 and TS0. The function m(E) represents unavoidable

    compliance costs associated with taxpaying, measured in labor units.

    Although evasion as a gamble is not explicitly treated in this model, Mayshar

    argues that it can be presented in this framework; to do so S is defined as that certain

  • 18

    payment which causes the same expected utility loss as the extra risk an evader takes on,

    for given expected tax payments. This forms the link between the A-S models of tax

    evasion and the models discussed in this section.17 From the perspective of an A-S

    evasion model, TS 0.

    Expression (7) looks familiar: the marginal rate of substitution between

    consumption and leisure equals the net wage. But note that the effective marginal tax

    rate, TX(X*,S*,E), permits more complex marginal tax rates than the standard linear tax

    model, where T(X*, S*, E) would equal tX*, and so TX would equal t. In expression (7),

    the effective marginal tax rate may depend on the sheltering activity of the taxpayer

    and/or the policy instruments of the government, interpreted more broadly than simply

    announcing a tax schedule. Expression (8) states that, because sheltering is accomplished

    by using labor, at an interior optimum its opportunity cost w(1-TX(⋅)) will be equal to its

    marginal private gain, which is the marginal tax saving, -TS.

    Slemrod (forthcoming) investigates a related model in which the private cost of

    17 Note that interpreting S, or C in the model of Slemrod (forthcoming) discussed below, as the risk bearingcost of evasion, will impose restrictions on the T(⋅) or C(⋅) functions.

  • 19

    achieving reductions in taxable income (denoted A, for income avoidance) is C(wL, A),

    where wL is true labor income; he argues that, in general, C10.18 If we imbed

    this avoidance technology into the taxpayer choice under a linear income tax, the

    maximization problem becomes

    (9) Max U(Y,L), L,A

    subject to Y=w(1 - L) - t(w (1 - L) - A) - C(wL,A).

    Before pondering the general implications of this formulation, first consider the

    special case where C(wL,A)=C(A). In this case the first-order condition for labor supply

    is identical to the standard model without avoidance. The first-order condition for A is

    simple and straightforward, C’ = t, implying that avoidance ought to be pursued until its

    marginal cost equals its marginal saving in tax liability, equal to t. In this situation a tax

    rate hike unambiguously increases A. Furthermore, its effect on L is no different than in

    the standard model, except to the extent that the income effect is altered by the possibility

    of avoidance.

    The story is enriched when the avoidance, or tax, technology becomes C(wL,A).

    The effective marginal return to working becomes w(1 - t - C1), where -wC1 is a subsidy

    to working that Slemrod (1994) dubs the "avoidance-facilitating" effect; for example, a

    given level of allegedly work-related deductions looks more plausible if it is taken against

    18 Slemrod (forthcoming) is less general than Mayshar in that it presumes a flat-rate statutory tax system; itdoes not presume that tax sheltering or avoidance must be "produced" with the taxpayer’s own time. Onesuperficial difference is the adoption by Slemrod of a cost function approach to avoidance, compared toMayshar’s production function for tax receipts.

  • 20

    a larger gross income. The term (t-C1) is analogous to TX in Mayshar’s model, and makes

    explicit how the avoidance technology influences the incentive to supply labor.

    Several insights emerge from this modeling of the tax environment. First of all,

    the substitution effect of labor supply does not respond identically to the two components

    of the statutory after-tax wage rate, w and (1-t). Changes in (1-t) trigger avoidance

    responses which are not triggered by changes in w. While both labor supply and

    avoidance respond to both w and (1-t), they do not do so symmetrically. This implies that

    econometric studies of labor supply (and avoidance) ought to differentiate responses to w

    and (1-t). Furthermore, one should not conclude, as does Rosen (1976), that a differential

    response to w and (1-t) necessarily represents "taxpayer illusion;"19 instead it could be

    reflecting the avoidance technology.

    Mayshar and Slemrod addressed the possibility that changes in the tax system will

    induce from taxpayers all three types of behavioral response. For example, an increase in

    the rate of a proportional income tax will provide an incentive to substitute leisure for

    goods, to (depending on the penalty structure) increase evasion, and increase avoidance.

    Other interactions among the three types of behavioral response have been investigated,

    as well. Cowell (1990a) develops a model in which the taxpayer can evade, but can also

    legally shelter income for a fixed cost Γ and a constant marginal cost γ, where γ < t.

    These cost assumptions generate the result that if an honest (or highly risk-averse) person

    shelters any of his income (Y), then he will automatically shelter all of it, and will do the

    latter if Γ + γY < tY. Cowell then investigates whether sheltering will co-exist with

    evasion, and asserts that the optimum is not characterized by an equality between the

    19 Although, note that in his empirical analysis Rosen (1976) does not detect a significant differentialresponse.

  • 21

    marginal cost of avoidance and evasion. This is because sheltering reveals to the tax

    authority that the taxpayer’s true income must be at least Γ /(t - γ). He argues that there

    may be a class of shelterers who would also have been evaders, had it not been for the

    attention drawn by sheltering, and that in some cases there may be a complete

    polarization between evaders and avoiders.

    In Cross and Shaw (1982), taxpayers must make expenditures to learn about and

    (in the case of avoidance) document both avoidance and evasion activities.20 Two

    avenues of interaction arise. First, in a progressive tax system, expenditure on avoidance

    or evasion reduces the marginal tax rate, thus reducing the return to engaging in the

    other.21 Second, investment in avoidance may reduce the marginal cost of evasion, or

    vice versa. For example, while investigating an illegal but undetectable "tax shelter," a

    (barely) legal tax shelter arrangement may be uncovered without much additional

    investment of time.


    4.a. The Extent of Tax Evasion

    4.a.1. Data problems

    Ascertaining the extent and characteristics of evasion immediately runs into two

    problems -- one conceptual and one empirical. The conceptual problem is that, although

    20 In some situations, more evasion may be associated with less cost. For example, not bothering to trace amiscellaneous source of income is less costly than tracking down whatever receipt or Form 1099 woulddocument the income. Not filing a return at all happens to minimize compliance cost.21 Alm (1988) also examines the simultaneous choices of evasion, avoidance and reported income, andinvestigates the effects of other fiscal variables on these choices.

  • 22

    one can assert that legality is the dividing line between evasion and avoidance, in practice

    the line is often blurry. Sometimes the law itself is unclear, sometimes it is clear but not

    known to the taxpayer, sometimes the law is clear but the administration effectively

    ignores a particular transaction or activity. The importance of these factors certainly

    differs across situations.

    The other difficulty is that, by its nature, tax evasion is not easy to measure --

    merely asking just won’t do. Several different approaches have been attempted. One

    approach relies on inferring the level or trends in noncompliance from data on measurable

    quantities, such as currency demand or national income and product accounts. The

    monetary indirect estimates are based on the presumption that most unreported economic

    activity takes place in cash, and that some time in the past the underground economy was

    small. In Gutmann (1977), increases in the ratio of currency to demand deposits since

    1937-41 measure the underground economy; in Feige (1979), changes since 1939 in the

    ratio of total dollar transactions to official GNP since 1939 measure it. Tanzi (1980)

    estimates regressions explaining the ratio of currency to money defined as M2, and

    interprets the portion explained by changes in the tax level as an indication of changes in

    the size of the underground economy. None of these approaches is likely to be reliable,

    however, as their accuracy depends either on unverifiable assumptions or on how well the

    demand for currency is estimated. The indirect noncompliance estimates based on

    discrepancies between national accounts measures of income and income reported to the

    tax authority are also problematic. For one thing, national income estimates of several

    key forms of income are based on tax return data. Second, there are many inconsistencies

    between how income is defined for tax purposes and for national accounts. However,

  • 23

    Engel and Hines (1999), in a study of tax evasion dynamics which focuses on the

    possibility of retrospective examination of previous-years' returns, study this measure of

    evasion in the U.S. for the years 1947 to 1993 and find that it responds as their model

    predicts. For example, annual fines and penalties imposed by the IRS subsequent to

    audits are correlated with contemporaneous and several lags of tax evasion as calculated

    from national income statistics.

    The most reliable source of information about tax compliance concerns the U.S.

    federal income tax, and exists because of the IRS’s Taxpayer Compliance Measurement

    Program, or TCMP. Under this program, approximately every three years from 1965

    until 1988 the IRS conducted a program of intensive audits on a large stratified random

    sample of tax returns, using the results to develop a formula used to inform the selection

    of returns for regular audits. The TCMP data consist of line-by-line information about

    what the taxpayer reported, and what the examiner concluded was correct. This data

    formed the basis for the IRS estimates of the aggregate "tax gap," and provides much

    useful information about the patterns of noncompliance with respect to such variables as

    income, occupation, line item, region of the country, age, and marital status. While

    informative, it is widely recognized that even the intensive TCMP audits imperfectly

    reveal particular kinds of noncompliance, such as income from the underground


    4.a.2 Patterns of noncompliance

    We cannot adequately review here what is known about the extent and nature of

    tax evasion for all taxes in all countries at all times. Rather, in what follows we offer a

  • 24

    few salient facts about the recent U.S. income tax, mostly gleaned from the TCMP data

    just discussed.

    1. With audit coverage hovering around 1% and an extensive information reporting

    and matching program, evasion is estimated to be 17% of true tax liability.22

    2. The extent of evasion varies widely across types of gross income and deductions;

    for example, the 1988 TCMP reports that the voluntary reporting percentage was 99.5%

    for wages and salaries, but only 41.4% for self-employment income (Schedule C). These

    percentages clearly correlate positively with the likelihood of income understatement

    being detected.

    3. Evasion (as measured by underreported income, not tax liability), rises with

    income but at a less than proportionate rate. Christian (1994) reports that in 1988,

    taxpayers with (auditor-adjusted) incomes over $100,000 on average reported 96.6

    percent of their true incomes to the IRS, compared to just 85.9 percent for those with

    incomes under $25,000.23

    4. Within any group defined by income, age, or other demographic category, there

    are some who evade, some who do not, and even some who overstate tax liability.24 For

    example, of middle-income (auditor-adjusted income between $50,000 and $100,000)

    taxpayers in 1988, 60% understated tax, 26% reported correctly, and 14% overstated tax.

    (Christian, 1994, p. 39).

    22 It is probably higher in most other countries. For example, Alm, Bahl, and Murray (1991) put the figure(for avoidance and evasion) at 46% for the Jamaican income tax of 1983. Richupan (1984) cites studies oftax evasion in developing countries indicating that it is not uncommon for half or more of potential incometax to be uncollected.23 One explanation for this pattern is almost certainly that tax returns of high-income households are morelikely to attract IRS attention. Another potentially important factor is that the TCMP results do not accountfor the noncompliance of business entities, which are more germane for higher-income individuals.24 Note that Erard (1997) concludes that a large fraction of noncompliant reports may be unintentional.

  • 25

    4.b. Determinants of Evasion

    Empirical attempts to more systematically establish how compliance responds to

    aspects of the tax environment have met with limited success, primarily due to the data

    problems discussed in Section 4.a.1.25 Three approaches dominate the literature.26

    4.b.1. Cross-sectional analysis

    Clotfelter (1983) was the first attempt to make use of the TCMP data to

    investigate how noncompliance responded to changes in the environment. He estimated a

    tobit model, explaining, for each of ten audit classes, noncompliance as a function of the

    combined federal and state marginal tax rate, after-tax auditor-adjusted income, and a set

    of demographic variables available on tax returns. The most striking conclusion is that

    noncompliance is strongly positively related to the marginal tax rate, with the elasticity

    ranging from 0.5 to over 3.0. This finding is apparently consistent with the basic A-S

    model, but not with the extension proposed by Yitzhaki.

    Beron, Tauchen and Witte (1992) investigate TCMP data aggregated by the IRS to

    the three-digit zip code level. They find that increasing the odds of an audit significantly

    increases reported AGI and tax liability for some, but not all, of the groups. In an attempt

    25 A former colleague, Harvey Galper, once put the problem this way: "Regression analysis of tax evasionis straightforward, except for two problems: you can’t measure the left-hand side variable, and you can’tmeasure the right-hand side variables!"26 In addition to the econometric methodologies discussed below, laboratory experiments typicallyinvolving students engaged in a multi-period reporting game, have been employed. (See, for example,Baldry (1987) and Alm, Jackson, and McKee (1992)). These results are subject to the canonical criticismsof laboratory studies: that the setting is artificial, and the participants are not demographically similar tothose making the actual decisions, and therefore do not come to the decision problems with the same arrayof experiences and expectations about the environment. These criticisms may be especially salient in thiscontext, because the experiments differ from general problems in choice under uncertainty only by thelabeling of the choice as having to do with taxes, and as compliant or not rather than gambling or not.

  • 26

    to deal with the potential endogeneity of the intensity of enforcement, they model the

    simultaneous determination of tax reporting and the log odds of an audit for each of the

    several audit classes in each zip code area. Their instrument for this is the level of IRS

    resources relative to the number of returns27. Although Beron, Tauchen, and Witte argue

    that it is a valid instrument because the IRS has not been able to distribute its resources

    among districts so as to achieve its goals, this is not convincing: it is reasonable that the

    IRS attempts to target its resources toward areas believed to be particularly noncompliant,

    thus invalidating use of IRS resources as an instrument.

    Subsequent studies have produced mixed results. Of particular interest is work by

    Feinstein (1991), who performed a pooled cross-section analysis of 1982 and 1985 TCMP

    data, thus mitigating the problem that in a single cross-section (other than for cross-state

    differences) the marginal tax rate is a (complicated, non-linear) function of income,

    making it difficult to separately identify the tax and income effect. Feinstein’s analysis

    suggests a negative impact of the marginal tax rate on evasion, which contradicts

    Clotfelter’s results but is consistent with the A-S model as adjusted by Yitzhaki.

    Klepper and Nagin (1989) investigate the characteristics of evasion across line

    items, and find that noncompliance rates are related to proxies for the traceability,

    deniability, and ambiguity of the items, which are in turn related to the probability that

    evasion will be detected and punished. They also find evidence of a "substitution effect"

    across line items, such that greater noncompliance on one item lowers the attractiveness

    27 Dubin and Wilde (1998) perform a similar analysis on the zip-code aggregated data, and use the sameinstrument. They defend this choice by claiming that, in an analysis of the time path of state-level IRSbudgets, they were found to be independent of compliance levels, and predominantly determined by theshare of total returns filed.

  • 27

    of noncompliance on others, because the latter jeopardizes the expected return to the

    former by increasing the probability of detection.

    4.b.2. Time-series analysis

    Dubin, Graetz and Wilde (1990) make use of state-level time series cross-section

    data from 1977 through 1986 to investigate the impact of audit rates and tax rates on tax

    compliance. They do not, though, have a direct measure of noncompliance, but instead

    use tax collections per return filed and returns filed per capita as (inverse) measures of

    noncompliance. They conclude that the continual decline in the audit rate over this

    period caused a significant decline in IRS collections -- amounting to $41 billion by 1985.

    4.b.3. Controlled experiments

    As discussed above, analysis of both cross-section and time-series historical data

    is subject to severe difficulties of measuring the parameters of the environment and in

    knowing the source of any variation in these parameters. Controlled experiments can

    avoid all of these problems, but, for cost and other implementation reasons, are rare.

    One recent exception is reported by Slemrod, Blumenthal, and Christian

    (forthcoming), in which the State of Minnesota Department of Revenue conducted a

    randomized controlled experiment with respect to four aspects of the tax compliance

    environment: the threat of an audit, the provision of special return preparation

    information services, moral appeals, and a redesigned tax form. With regard to the first,

    they find that, for low- and middle-income taxpayers, a threat of certain audit28 produced

    a small, but statistically significant, increase in reported income, which was larger for

    28 The audit threat was delivered by letter in January following the tax year.

  • 28

    those with greater opportunities to evade.29 However, for high-income taxpayers the

    audit threat was associated with on average a lower income report. The authors speculate

    that sophisticated, high-income, taxpayers view an audit as a negotiation, and view

    reported taxable income as the opening (low) bid in a negotiation which does not

    necessarily resulted in the determination and penalization of all noncompliance. Based

    on the same experiment, Blumenthal, Christian and Slemrod (1999) find no evidence that

    either of two written appeals to taxpayers' consciences had a significant effect on

    aggregate compliance.


    5.a. Dimensions of Avoidance

    Stiglitz (1985) distinguishes three basic principles of tax avoidance within an

    income tax: postponement of taxes, tax arbitrage across individuals facing different tax

    brackets (or the same individuals facing different marginal tax rates at different times),

    and tax arbitrage across income streams facing different tax treatment. Many tax

    avoidance devices involve a combination of these three principles. In an example used by

    Stiglitz, the basic feature of an Individual Retirement Account (IRA) is the postponement

    of tax liability until retirement; if the individual faces a lower tax rate at retirement than at

    the time the income is earned, then the IRA also features tax arbitrage between different

    rates. Finally, if the individual can borrow to deposit funds in an IRA and the interest

    incurred to finance the deposit is tax deductible, then the IRA is a tax arbitrage between

    29 The approach is a "difference-in-difference" analysis; that is, the increase in reported income over theprevious year of the treatment group is compared to the increase in reported income of the control group.

  • 29

    two forms of capital, one of which is taxed, and the other of which is not taxed.30 Stiglitz

    argues that, with perfect capital markets, these three principles can be exploited to

    eliminate all taxes while leaving the individual’s consumption and bequests unchanged

    relative to the zero tax case, and facing no more risk than in the original situation. But

    capital markets are not perfect, and therefore all tax liability is not eliminated by tax

    avoidance,31 and to reduce tax liabilities distorting actions (such as investment in sectors

    where it is easier to convert ordinary income into capital gains) are utilized. There is

    considerable empirical evidence testifying to the extent and tax sensitivity of these kinds

    of avoidance behavior.

    5.b.1. Retiming

    There is abundant support for the notion that the timing of certain transactions can

    be extraordinarily responsive to changes in tax rates. Perhaps the most striking example

    was the response of capital gains realizations to the tax rate increase scheduled to occur

    on January 1, 1987, but fully anticipated by the fall of 1986. Aggregate realizations in

    1986 were twice what they were in any previous year or for several years thereafter. As

    Burman, Clausing, and O’Hare (1994) document, capital gain realizations on corporate

    stock in December of 1986 were seven times higher than in the previous December.

    Another striking example of timing response is provided by Goolsbee (forthcoming), who

    documents that, in advance of the expected 1993 increase in the U. S. top individual tax

    30 The IRA example makes clear that in certain cases (some of) the avoidance behavior is the result of aconscious tax policy choice, in this case with the intent of increasing saving. Another excellent example iscapital gains, where taxation upon realization rather than accrual allows for deferral of tax liability, ofteninto periods of lower taxation, and where gains are completely excused from taxation at death due to thestep-up of tax basis.31 There are also policy responses to avoidance, such as limits on loss offsets and interest deductions.

  • 30

    rate, corporate executives realized a huge amount of income in 1992, primarily through

    exercising non-qualified32 stock options.

    Sophisticated econometric techniques using panel data have been developed for

    separately identifying the timing responses to tax rate changes over time from the

    permanent behavioral response to a changed tax rate. These new techniques have been

    applied to both capital gains realizations (Burman and Randolph, 1994) and charitable

    contributions (Randolph, 1995). In both cases the results suggest that the retiming effect

    dominates the permanent effect.

    5.b.2. Tax arbitrage

    Tax arbitrage activity takes advantage of inconsistencies in the tax law, featuring

    economically offsetting positions which have asymmetric tax treatments. Examples

    range from sophisticated derivative financial instruments to the more mundane cases of

    doing tax-deductible borrowing to finance tax-deferred IRA contributions or tax-exempt

    bond purchases.

    5.b.3. The classification of income

    The classic example of income reclassification, also termed income shifting, is

    turning ordinary capital or labor income into preferentially-taxed capital gains. In another

    example, Maki (1996) and Scholz (1994) have documented that, following the Tax

    Reform Act of 1986, there was a large shift from no-longer-deductible consumer interest

    into still-deductible mortgage or home equity loans. There is anecdotal evidence that,

    32 For non-qualified stock options, the difference between the exercise price and the issue price is taxable atordinary income tax rates at the time of exercise, and is deductible from the employer's taxable income atthe same time.

  • 31

    following the introduction of the R&D credit in the United States, much business activity

    was "discovered" to have a significant research component. Gordon and MacKie-Mason

    (1991, 1997) have investigated how, when the Tax Reform Act of 1986 lowered the top

    individual rate below that of the corporate rate, there was a large shift from C

    corporations into S corporations, which are taxed like partnerships and therefore are not

    subject to the corporation income tax. Gordon and Slemrod (forthcoming) discuss the

    shifting of income between the corporate and individual tax base via the method of

    compensation, and document evidence of such shifting in the United States.

    5.c. The Extent of Avoidance

    No one has attempted to calculate for avoidance a counterpart to the aggregate

    evasion "tax gap." There is, though, some indirect evidence that the avoidance tax gap is

    large. Gordon and Slemrod (1988) calculated that the U.S. tax system of 1983 raised

    approximately zero revenue from taxing capital income, due to the combination of

    legislated deviations from a pure income tax and tax arbitrage.33 As to the incidence of

    the avoidance opportunities, Agell and Persson (1988) and Gordon and Slemrod (1988)

    argue that the availability of tax arbitrage opportunities will generally benefit those at the

    bottom and top of the tax rate distribution, to the disadvantage of those in the middle.

    This generally corresponds to low- and high-income individuals, respectively, but there

    are exceptions to that rule; high-income individuals benefit through their ownership of

    tax-preferred pension assets.

    33 It is likely that the Tax Reform Act of 1986 mitigated the avoidance tax gap by reducing the dispersion ofmarginal tax rates and tightening up the rules about tax arbitrage behavior.

  • 32


    Having completed a review of the positive, or descriptive, analysis of tax evasion and

    avoidance, we turn now to the normative analysis of taxation. However, before we

    proceed to that task, we must first reconsider the fundamental building blocks of tax

    analysis -- the evaluative criteria of equity and efficiency -- to check whether these

    concepts need to be revised.

    6.a. Equity

    6.a.1. Vertical equity

    Analyses of the distributional impact of taxation, especially those based on tax

    return data, ought to account for the presence of evasion. The evidence cited in Section

    4.a.--that noncompliance as a fraction of true income declines with true income--suggests

    that standard analyses of incidence based on the statutory rates and base may understate

    the progressivity of the tax burden;34 Bishop, Chow, Formby, and Ho (1994) find this for

    the United States using the 1985 TCMP data, although Alm, Bahl, and Murray (1991)

    reach the opposite conclusion about Jamaica.35

    6.a.2. Horizontal equity

    Horizontal equity -- the idea that equals should be treated equally by the tax

    system, or that tax liability should not depend on any of a set of irrelevant characteristics -

    - is central to an assessment of the impact of tax avoidance and evasion. To see this,

    34 This evidence also suggests that tax return data may overstate the inequality in the distribution ofincomes. Because the data on tax evasion are flawed, one should keep in mind that the theoreticalarguments discussed in Section 5.c. imply that data on reported incomes will understate the true dispersionof income.35 A complete incidence analysis would account for the costs borne by evaders in the form of exposure torisk and concealment expenses, neither of which is accounted for in the studies mentioned.

  • 33

    compare two tax situations, one in which there is a linear income tax rate of 20% and

    everyone reports their true income, and another in which the tax rate is 40% and everyone

    (costlessly) reports exactly half their income. In this case the two systems are identical

    with respect to both horizontal and vertical equity. Now imagine that, in the second

    system, on average everyone reports half their income, but that the fraction differs

    systematically by income. In that case replicating the progressivity of the first tax system

    will require a more complicated, non-linear, system of rates. If, however, evasion varies

    within income classes, no revision of the tax rate schedule can compensate, and there will

    be horizontal inequity.

    In the context of the A-S model of tax evasion, the horizontally inequitable tax

    burden will depend on the taxpayer’s degree of risk aversion. Less risk-averse

    households will gain more from the availability of a gamble with given positive expected

    value. In contrast, common parlance would ascribe any horizontal inequity to variations

    in honesty, with the honest, or dutiful, citizens left holding the bag by the dishonest. In

    the typical economic model, though, there are no honest or dishonest individuals, only

    utility-maximizers; thus, this distinction can be introduced only artificially by simply

    positing that some individuals do not pursue tax evasion. The same kind of artificial

    differentiation across people can be made with regard to tax avoidance by positing that

    some people have an aversion to such behavior; as Steuerle (1985, p. 78) says: "Some

    taxpayers simply do not enjoy playing games no matter what the certainty of the return;

    the U.S. tax system is designed to insure that such individuals pay a greater share of the

    tax burden than those who are not so hesitant." Steuerle (p. 19) concludes that "taxpayers

  • 34

    pay unnecessary taxes because of the simplicity of their filing response or their lack of

    knowledge of the tax laws."

    6.a.3. Incidence

    The theory of tax incidence -- who bears the burden of a given tax structure --

    begins with three basic principles: (i) the burden of all taxes must be traced back to

    individuals; (ii) individuals with relatively elastic demand (or supply) of a taxed good

    tend to escape the burden of tax imposed on that good; and (iii) in the long run the

    incidence of a tax levy does not depend on which side of the market bears the legal

    responsibility for remitting the tax to the government. Introducing avoidance and evasion

    preserves the methodological importance of the first two principles,36 but calls the third

    into question. A complete analysis of the incidence of a particular tax requires specifying

    the remittance process and positing an avoidance technology for both the suppliers and

    demanders of the taxed good.

    Avoidance opportunities alter the analysis of incidence for two separate reasons.

    First, their presence affects the behavioral response to a change in the tax system, and this

    alters what otherwise would be the change in equilibrium prices. Second, the presence of

    avoidance alters the link between tax-inclusive prices and welfare. This suggests that the

    incidence (not to mention the efficiency) of a tax may depend on which side of the market

    the responsibility for remittance falls. That is in stark contrast to the standard model,

    under which that is irrelevant to the long-run incidence.37

    36 Section 2.d discusses some models of tracing the incidence of tax evasion.37 There are exceptions. Consider, for example, the debate between Tanzi (1992) and Dixit (1991) over theimplications of tax collection lags for the optimal amount of inflationary finance. Tanzi (1992) argues that,when consumption taxes are collected by firms in advance and held by them for the duration of thecollection lag, inflationary finance implies a real redistribution of income from consumers to sellers.

  • 35

    6.a.4. Are changes in the social welfare function necessary?

    In models with heterogeneous citizens, the standard objective function is a social

    welfare function which has as arguments the utility level of each citizen -- accepting the

    individuals’ own relative valuations of goods and services -- where the shape of the social

    welfare function implicitly determines the social value placed on the distribution of

    utilities as opposed to the sum of utilities. In the presence of uncertainty, the expected

    utilities of individuals are the relevant arguments -- accepting the risk preferences of

    consumers. Cowell (1990b) questions the appropriateness of according the same social

    weight to investigated and guilty taxpayers as is applied to the innocent or uninvestigated,

    and argues that there may be a case for putting a specific discount on the utility of those

    "who are known to be antisocial" (p. 136). Cowell investigates a few alternative social

    objective functions, including one in which any private benefit derived from the proceeds

    of evasion is assigned a social weight of zero, but in our opinion no convincing

    alternative that provides reasonable policy prescriptions has yet been presented.

    6.b. A Taxonomy of Efficiency Costs

    In the standard model the efficiency cost of taxation is entirely due to the fact that,

    because of the change in relative prices, individuals are induced to select socially

    suboptimal consumption baskets -- to substitute away from relatively highly-taxed goods

    to relatively lightly-taxed goods, such as leisure. A standard exercise in optimal taxation

    theory is to describe the tax system that minimizes these costs, or to describe the tradeoff

    between these costs and the distribution of welfare in the society.

    In the presence of avoidance and evasion, a broader concept of efficiency cost is

    needed. In what follows, we describe and comment on three additional components of

  • 36

    the social cost of taxation and discuss the problems that arise in introducing these costs

    into formal models of optimal taxation.

    6.b.1. Administrative costs

    Tax administrations deal, among other things, with information gathering. But

    this is a difficult element to model because information varies in quality. There is a

    qualitative difference between an auditor "knowing" that a given taxpayer is evading and

    having sufficient evidence to sustain a court finding to that extent. Also, the cost of

    gathering information depends on how accessible the information is, and whether it can

    be easily hidden. There are several advantages to taxing a market transaction relative to

    taxing an activity of the individual such as self-consumption. First, in any market

    transaction there are two parties with conflicting interests. Hence, any transaction has the

    potential of being reported to the authorities by one unsatisfied party. A second property

    is that the more documented the transaction, the lower is the cost of gathering information

    on it. For this reason it is easier to tax a transaction that involves a large company, which

    needs the documentation for its own purposes, than to tax a small business, which may

    not require the same level of documentation. Finally, market transactions establish arms-

    length prices, which greatly facilitate valuing the transaction. Administrative cost may

    also be a function of the physical size and the mobility of the tax base (it is harder to tax

    diamonds than windows), whether there is a registration of the tax base (e.g., owners of

    cars, holders of drivers’ licenses), the number of taxpayer units, and information sharing

  • 37

    with other agencies.38 It is also an increasing function of the complexity and lack of

    clarity of the tax law.

    Administrative costs possess two additional properties that complicate the

    modeling of tax administration issues: they tend to be discontinuous and to have

    decreasing average costs with respect to the tax rate. To see the first property, consider

    two commodity tax rates, denoted by t1 and t2. If t1 = t2, then only the total sales of the

    two commodities need be reported and monitored. If, however, the two rates differ even

    slightly, then the sales of the two commodities must be reported separately, doubling the

    required flow of information. There are decreasing average costs because the cost of

    inspecting a tax base does not depend on the tax rate (except to the extent that people are

    more inclined to cheat with a higher tax rate). Hence, a higher tax rate reduces the

    administrative cost per dollar of revenue collected (Sandford, 1973). Administrative cost

    may also be a function of the combination of the taxes employed and their rates, because

    the collection of information concerning one tax may facilitate the collection of another

    tax (e.g., inspection of VAT receipts may aid the collection of income tax).

    6.b.2. Compliance costs

    Slemrod (1996a) estimates that, for the U.S. income tax, the private compliance

    cost is about 10 cents per dollar collected. Sandford (1995) presents estimates for a

    variety of taxes in several countries. Some of that cost is an unavoidable cost of

    complying with the law, and some of it is voluntarily undertaken in an effort to reduce

    one’s tax bill, but in either case it approximately represents resource costs to society. In

    38 A good description of the properties of administrative cost can be found in Shoup, Blough, andNewcomer (1937), pp. 337-51.

  • 38

    almost all cases the private compliance costs dwarf the public administrative costs of

    collecting taxes, which the IRS estimates at 0.6 cents per dollar collected for all the taxes

    it administers.

    Integrating compliance costs into formal models in a meaningful way is tricky.

    As an example of the modeling difficulties this topic poses, consider the following

    problem: when is it optimal to delegate to employers the authority to collect taxes and

    convey information about employees, thus requiring the administration to audit both the

    taxpayer agent and the taxpayer himself, and when is it optimal to deal only with the

    employee? Clearly, given that the employer already has the necessary information, it

    would save administrative costs to require him to pass it along to the tax administrator.

    This might also reduce total social costs if the cost of gathering information by the

    administration is higher than the increase in cost caused by imposing a two-stage

    gathering system.39

    However, the potential efficiency of involving taxpayers in the administrative

    process must be tempered with a practical consideration. Administrative costs must pass

    through a budgeting process, while compliance costs are hidden. Hence, there may be a

    tendency to view a policy which reduces administrative cost at the expense of an equal

    (or greater) increase in compliance costs as a decrease in social cost, because it results in

    a decrease in government expenditures. We will discuss this issue further in Section 7.

    39 Note that a withholding system requires two information gathering systems and might generate incentivesfor the withholding agent to evade the taxes it collects, or to collaborate with withholdees in withholdingless than required (Yaniv, 1988 and 1992). In a period of rapid inflation, the gain to the agent fromwithholding may exceed the cost.

  • 39

    6.b.3. The risk-bearing costs of tax evasion

    In the Allingham-Sandmo model, tax evasion occurs only if the taxpayer expects

    to increase his expected income by evading taxes, including the expected fines that he

    would have to pay if he were caught; it continues until, at the margin, the increased

    expected income is offset by the increased risk-bearing. Hence, a taxpayer who evades

    taxes increases both his exposure to risk and his expected income. This additional

    exposure to risk is a deadweight loss to society. In principle, the taxpayer could be better

    off under an agreement whereby the taxpayer pays at least as much as the government

    currently collects, while the government ceases to audit. Assuming a risk-neutral

    government, the risk-bearing cost of tax evasion is equal to the risk premium that the

    taxpayer would be ready to pay in order to eliminate the exposure to risk (Yitzhaki

    (1987)). Depending on the other assumptions about the probability of detection, the

    penalty structure, and risk aversion, the risk-bearing costs of evasion may be a continuous

    function that increases with the tax rates. These costs are in addition to the compliance

    costs voluntarily incurred by an individual attempting to minimize the expected cost by

    camouflaging the evasion or shifting to an otherwise less remunerative occupation.


    7.a. Optimal Tax Administration and Enforcement

    Avoidance and evasion pose two challenges for the normative analysis of

    taxation. First, they introduce a new set of policy instruments whose optimal setting is at

    issue. These include the extent of audit coverage, the penalty imposed on detected

    evasion, and the structural integrity of the tax code itself, which determines the extent and

  • 40

    nature of avoidance opportunities. Second, they invite a rethinking of standard taxation

    problems, including the optimal setting of commodity tax rates and optimal progressivity.

    7.a.1. Optimal Penalties

    Consider the A-S model of a representative consumer whose true income is

    exogenous and whose only choice concerns how much of that income to report. This

    choice depends on two policy instruments set by the government, p, which has a resource

    cost due to the need for auditors and the related infrastructure, and θ, which is a fine for

    detected evasion, which is a transfer with no resource cost.

    It has been well known since Becker (1968) that in this setting a government

    concerned with maximizing the ex ante utility of its representative citizen will want to set

    θ as high as possible, allowing p to be as low as possible. This policy of "hanging

    violators with a probability of zero" deters evasion while minimizing the resource cost of

    the deterrent -- p represents a real resource cost but θ is simply a transfer. But this kind

    of model ignores, inter alia, the possibility of a corrupt tax administrator who abuses the

    system or, alternatively, harshly punishes someone who commits an honest mistake.40

    The harsher the penalty, the more damage that can be inflicted by a corrupt administrator

    or, in the case of an honest mistake, the more capricious the system is. Hence, the harsher

    the penalty, the more detailed and cautious the prosecution process should be, although

    this may increase its administrative costs. In the absence of modeling the interaction

    between the penalty rate and administrative costs, analytical models usually assume a

    ceiling on the penalty rate.

    40 Polinsky and Shavell (1999) examine this and other issues involved in the optimal setting of penalties forcrime including but not restricted to tax evasion.

  • 41

    7.a.2 Optimal randomness

    Auditing some taxpayers and not others inevitably introduces some ex ante

    uncertainty and some ex post horizontal inequity. This suggests a link to an earlier

    literature in public finance, in which Stiglitz (1982) and Weiss (1976) each argued that,

    even in a world of risk-averse citizens, it may be optimal for the government to introduce

    some randomness into its net tax (or transfer) to individuals. The argument depended on

    the second-best nature of the problem, in which an income tax distorted the labor-leisure

    choice. For some utility functions, Stiglitz and Weiss argued, the introduction of random

    payments induced people to work harder, thus mitigating the labor market distortion; in

    some cases the value of the increased labor more than offset the utility loss from the

    randomness introduced.

    This argument has clear implications for the optimal enforcement of the income

    tax, because it suggests that one of the presumed social benefits of greater enforcement --

    the reduced uncertainty of payment of a given expected value of taxes -- may be mitigated

    by the increased labor supply distortion. Weiss uses approximations around the point of

    no evasion to describe the condition under which allowing some degree of evasion can

    both increase revenue and increase welfare. However, Yitzhaki (1987) shows that, in the

    examples used by Weiss, the condition that allows successful evasion is identical to the

    condition that the solution is on the declining portion of the Laffer curve; in this case, any

    reduction of the tax rate would increase welfare and increase revenue. This suggests that

    the improvement was not caused by allowing evasion. We conclude that neither the

    practical nor hypothetical relevance of this point has not yet been demonstrated.

  • 42

    7.a.3. The optimal extent of enforcement

    For a given penalty structure how much resources should be devoted to enforcing

    the tax laws? Or, in other words, what is the optimal probability of detection, p? Many

    widely-used textbooks and several IRS commissioners presume that the answer is to

    increase p until the marginal increase of revenue thus generated raised equals the

    marginal resource cost of so doing. As Slemrod and Yitzhaki (1987) show, however, this

    rule is incorrect. Intuitively, although the cost of increasing p (hiring more auditors,

    buying better computers, etc.) is a true resource cost, the revenue brought in (through

    assessed fines as well as higher compliance) does not represent a net gain to the economy,

    but rather a transfer from private citizens to the government. The correct rule equates the

    marginal social benefit of reduced evasion to the marginal resource cost; the social cost is

    not well measured by the increased revenue, but is in this model related to the reduced

    risk bearing that comes with reduced evasion.41 This result implies that privatization of

    revenue collection will inevitably lead to a socially excessive amount of resources

    devoted to that purpose unless restrictions are put on the resources and behavior of the


    7.a.4. Optimal auditing rules

    One of the key simplifying assumptions of the Allingham-Sandmo model is that

    the probability of evasion being detected is fixed and unrelated to any actions of the

    taxpayer. In Section 2.b we investigated the implication of p increasing with the amount

    of evasion, but this relationship was exogenously imposed. Other models allow the audit

    41 Note, though, that Baldry (1984) has shown that complete enforcement of income tax laws (p highenough to deter evasion) is inefficient.

  • 43

    strategy of the tax collection agency (henceforth the IRS) to depend on the report of the

    taxpayer in a way that maximizes an explicit objective function; the taxpayer, in turn,

    forms some expectation of what the IRS’ auditing rule is, and acts accordingly. In

    modeling the game between taxpayers and the IRS, researchers have generally assumed

    that the IRS attempts to maximize net revenue collected. As we discussed earlier, this is

    not likely to characterize the social-welfare-maximizing solution to how big the

    enforcement budget ought to be, although it might characterize the optimal allocation of

    resources for a given IRS budget. Another critical model element is whether it is

    assumed that the IRS can commit to an announced audit rule, or whether it cannot

    commit, and therefore will opportunistically audit whatever returns it wishes once the

    returns are filed. Finally, it is critical whether the IRS budget is assumed to be fixed.

    Following Reinganum and Wilde (1985), models of this question generally assume

    that the probability of audit depends on reported income only. Most papers conclude that

    the optimal strategy in this context is to randomly audit individuals who report below

    some threshold level of income. In equilibrium only low-income individuals report

    honestly, while high-income taxpayers report exactly at the threshold level of income and

    are never audited. Sanchez and Sobel (1993) derive this result in the context of risk-

    neutral taxpayers with a continuous distribution of actual income and no labor supply

    decisions, and where penalties for detected evasion are bounded and exogenously set.

    Cremer and Gahvari (1996) reach similar conclusions when they allow for endogenous

    labor supply, although they consider just two types of individuals. Mookherjee and P'ng

    (1989) consider risk-averse individuals. Imposing mild restrictions on the level of risk

    aversion, they show that the optimal policy is characterized by random audits and finite

  • 44

    penalties. It is still true that above some income level taxpayers are not audited, but it is

    no longer true that everyone reporting an income below that level is honest.42 Scotchmer

    (1987) relaxes the assumption that the IRS can only observe the taxpayer's report, and

    instead assumes that it is possible to assign taxpayers to a number of audit classes based

    on observable characteristics. Although the optimal audit policy within each class is

    similar to that described above, this policy introduces a regressive bias to the effective tax

    system, because the agency will audit taxpayers with low-income reports with higher

    probability that high-report taxpayers, thus making it less attractive for low-income

    taxpayers to underreport income. This bias may be difficult to undo through the statutory

    tax system if the tax code cannot depend on the audit class.

    This state of affairs provides an obvious temptation to the IRS to reverse its pre-

    announced audit rule and